Sync ingest schema + wire sentence_pair_classification & embeddings into CLI category registry

[saadqbal]

Problem

The Schema drift check CI job (scripts/sync-schema.sh --check) is failing on every open PR (including #160), because internal/schema/ingest.v1.json has drifted from upstream tracebloc/data-ingestors@master. Upstream added:

• two new task categories: sentence_pair_classification, embeddings
• a content_hash data_id strategy
• related description updates

The drift predates and is unrelated to #160 (it's stale on develop too).

Why it's not just a schema bump

Running scripts/sync-schema.sh alone breaks TestRegistryCoversSchemaCategories: the CLI's authoritative categoryRegistry (internal/push/category.go) must cover every schema category, and the two new ones aren't wired in.

Plan

• scripts/sync-schema.sh to vendor the current upstream schema.
• Add sentence_pair_classification + embeddings to categoryRegistry as FamilyText (both stage raw .txt from texts/ per data-ingestors conventions.py TEXT_CATEGORIES).
• Mark both CLISupported: false with an UnsupportedNote (conservative default, mirroring causal_language_modeling/seq2seq/token_classification) until the tracebloc ingest discover/build staging path for them is confirmed. Reviewer: flip to true if the push path already supports them.
• Update the pinned lists in category_registry_test.go.

Acceptance

• scripts/sync-schema.sh --check passes.
• go build/vet/test ./... green.
• Unblocks the schema-drift check across open PRs once merged to develop.

[LukasWodka]

Shipped in #163 (merged to develop, c0344bb). sentence_pair_classification + embeddings are wired into the category registry as FamilyText / CLISupported:false with an UnsupportedNote, the content_hash data_id strategy is vendored into ingest.v1.json, and the registry⇄schema parity test pins them. Schema-drift check green. Closing as done. (Flipping those two to CLISupported is separate future work — see #105 for the analogous texts-layout support.)